Micro Plastic In Water Detection Using Deep Learning Approach

Abstract

The growing occurrence of micro-plastics in the water bodies is a major concern to environmental and human health. Historical methods of detection have been time consuming, costly and have used special laboratory set ups. In this project, the author has submitted a proposal built using deep learning that utilizes image classification and detection micro-plastics in water. The populated dataset was constructed and pre-processed (by resizing, contrast optimizing and gamma correction) in the form of labeled image of both clean water and water contaminated with micro-plastics (1,554 images, between which there are equal numbers). A few models of deep learning were trained and tested such as VGG19, ResNet50, Xception, DenseNet21, and AlexNet. Of them, the Xception model had the best accuracy of 97 per cent and was deployed. Streamlig was used to create a user-friendly interface where it is possible to upload an image in real-time and detect it. The findings show the possibility of deep learning in the automation of micro-plastics detection, and can serve as an accessible and scalable environmental monitoring and research tool